Research On Feature Optimization Algorithm Based On Sparse Representation And Diagnostic Application Of Lung Cancer

With the rapid development of information technology,high-dimensional data will inevitably appear in many fields,in the field of pattern recognition,because the improvement of the recognition accuracy rate depends on the comprehensive and effective feature data,therefore,high-dimensional features often appear,which in turn causes so-called "dimensional disasters." In order to avoid the "dimensional disaster" problem,it is necessary to reduce the dimensions of high-dimensional feature data.Data dimension reduction is to optimize the high-dimensional feature data so as to reduce the data dimension,remove redundant information as much as possible,and retain valid information,which minimizes the loss of information after dimension reduction.Although the current feature optimization algorithms can basically meet the requirements of dimensionality reduction,these methods can hardly find inherent low-dimensional information hidden in high-dimensional data.Therefore,the feature optimization algorithms still have problems that are difficult to be applied in the field of pattern recognition.Based on the above situation,this paper presents a feature optimization algorithm based on improved sparse representation,and applies this feature optimization al gorithm in the diagnosis of benign and malignant lung nodules.This article mainly completed the following aspects:(1)In this paper,we have studied the algorithm of dimension reduction for a large number of feature data,analyzed the existing problems of dimensionality reduction algorithms,detailed the related theories of sparse representation,it reveals the advantages of sparse representation algorithms over existing data dimensionality reduction algorithms.(2)This article analyzes the existing sparse representation methods.Among them,the typical K-SVD dictionary learning algorithm has been widely used due to its good sparse representation effect,but its learning of the internal features of high-dimensional data is still not enough,so that the effect of sparse representation is not optimal.(3)This paper improves the sparse representation K-SVD dictionary learning algorithm.In a typical K-SVD algorithm,the OMP matching pursuit algorithm is usually used in the sparse decomposition stage.Here,we use an improved CoSpOMP algorithm for sparse decomposition,this algorithm is more efficient than the typical OMP algorithm to reuse coefficients and converge faster.In the dictionary update phase,the initial dictionary is usually a dictionary directly formed by randomly selecting the signal arrangement.In this paper,the discrete cosine matrix is selected as the initial dictionary because the signal energy after DCT processing is very concentrated.Simulation experiments show that the improved algorithm has better sparse representation effect and stronger convergence.(4)The feature optimization algorithm based on improved sparse representation proposed in this paper is applied to the diagnosis of benign and malignant pulmonary nodule.In view of the fact that the feature information resulting from the selection of single-layer maximum area ROI feature in existing diagnostic techniques is not comprehensive enough,this paper first extracts the ROI high-dimensional feature data of multi-layer slices of pulmonary nodules and then sparsely expresses the high-dimensional feature data.Finally,the sparsely processed data sets are classified by support vector machine.The experimental results show that compared with the existing methods,the method proposed in this paper can improve the accuracy,sensitivity and specificity of benign and malignant pulmonary nodule diagnosis while ensuring the diagnostic efficiency,and the improved K-SVD sparse representation feature optimization algorithm has better diagnostic performance than the traditional K-SVD,which illustrates the effectiveness of the proposed algorithm.